Good Morning! Here's what I have for you in today's newsletter:
xAI releases Grok 4.7, longer reasoning at Grok 4.6's price
Google finishes Gemini Notebook Live Chat rollout to Ultra on mobile
Perplexity Computer generates videos with MiniMax H3 and Seedance 2.5
Turn Grok Bot into five hires that run your company for you
4 new AI tools worth trying today
AI TOOLS
xAI released Grok 4.7, built on a larger base model with a longer reinforcement learning run, served at Grok 4.6's price and speed in the API, Cursor, and Grok Build.
Training was weighted toward problems that take many hours, so the model verifies its own output before returning it and manages long context across a 500K token window.
Coding scores rose to 46.3 percent on CursorBench 4.0 from 40.4 for Grok 4.6, and EEBench, an electrical engineering test, jumped to 64.0 percent from 53.0.
A rebuilt safeguard stack lets through 3.3 percent of risky cyber prompts on HackerBench v0.3 and scores 62.4 percent on LatchBio's biosafety benchmark, with red-team access for security partners.
If you already run Grok 4.6 in Cursor, Grok Build, or the API, switching to 4.7 costs nothing and gets you a model that stays on long tasks and double-checks itself. The price to watch is tokens, since longer reasoning can burn more of them per job. It is also free inside Grok Build, so test it there first.
AI MODELS
Google completed the rollout of Live Chat in Gemini Notebook, formerly NotebookLM, to every Google AI Ultra subscriber on Android and iOS, so users can talk to a notebook and get spoken answers drawn only from the documents inside it.
Users speak questions about their uploaded documents in nearly 100 languages, and answers stay tied to those sources, with no content pulled from the open web.
The notebook can be interrupted mid-sentence to change direction or ask a follow-up, and the conversation is saved to chat history for later review.
Live Chat runs on the Gemini 3.8 Live audio models, replies in English only, and is limited to users 18 and over, with Google AI Pro and Plus access to follow.
Studying from a notebook of lecture PDFs becomes a spoken back-and-forth on a phone, so you can quiz yourself on the walk to class or get a concept re-explained while cooking. Because the answers only draw on your own sources, it works as a tutor for exactly the material you uploaded.
AI MODELS
Perplexity added video generation to Computer, its agent for Pro and Max subscribers, using MiniMax H3 and ByteDance Seedance 2.5 to produce finished clips inside the same thread as a project.
A single request for a campaign clip, product demo, or social asset returns the video alongside the copy and creative Computer already drafted for it.
Both MiniMax H3 and Seedance 2.5 are available as backends, and Computer handles the model call itself, so the user only describes the outcome.
Video output sits with Computer's existing research, writing, and browsing work, so a launch plan, its landing page copy, and its promo clip can come from one conversation.
The campaign copy and the video that goes with it come out of the same chat, with no second tool or subscription to open. For a marketer who needs a 15-second product clip by lunch, Computer makes the clip itself.
HOW TO AI
Running solo means everything waits for you. Grok Bot changes that. Each bot gets its own cloud computer, signs into your tools the way you do, and keeps working after you close your laptop. One account holds up to 50 bots, sharing files, browser sessions, and logins.
Step 1: Get access
Download the desktop app at x.ai/bot, sign in with the account that owns your plan. During onboarding it asks which tools you use daily, answer specifically, it shapes what gets suggested later.
Step 2: Hire the first bot
A prompt is a request. A bot is a role. A charter has three parts: what it owns, what good output looks like, and what it must never do without asking first.
You are my Outbound SDR.
// what you own
Run daily outbound without asking me.
Pull 40 prospects matching [industry, size, title, region].
Research each one. Write a 3-line cold email, send from [inbox], book from my calendar when someone says yes.
Send me a 5-line summary each morning.
// where you stop
Never buy a list. If a reply mentions legal, money, or contract terms, stop and ask me.Test it immediately with something that should trip the "where you stop" line, and confirm it stops instead of proceeding.
Step 3: Connect the tools
Settings → Plugins. Connections are shared across your entire account, connect Gmail once, every bot can use it. For anything without a plugin, the bot hands you the login screen, you authenticate, it resumes. You never type a credential into a chat message.
Step 4: Teach and schedule it
Open the bot's cloud computer, click Teach a task, walk through the workflow while it watches, click Stop. It saves the flow as a named, reusable skill.
Every weekday at 7:00 AM, check my calendar, inbox, and #launches Slack. Give me one short brief. Do not send anything, just prepare and report.Step 5: Add a chief of staff
Put multiple bots in one group chat and they coordinate without you.
@Mara gather 40 qualified prospects. Research each one and flag the top 10.
@Cole take Mara's top 10 and draft personalized outreach. Draft only, do not send.Create a chief-of-staff bot whose job is coordinating the others:
Look at the bots I have created, these are your team. Tell me if we need to change anything or add more bots.
P.S. You can access all the AI courses and trainings (including the full version of this one), prompts and workflows if you upgrade.
Xiaomi released MiMo-V2.6 Pro and Flash, two open-weight omnimodal models trained with scaled reinforcement learning, where Pro scores 46 on the Artificial Analysis Intelligence Index, the highest of any open model, with training code and RL environments published.
ElevenLabs upgraded Studio to 4.0 inside ElevenCreative, so video, images, voice, music, and sound effects are generated inside the project itself, then edited, captioned, and exported without leaving the editor.
Lovable added Jev to the Lovable AI Gateway, so builders can wire AI-powered features into their apps using Typesafe's model, free of charge through Sunday, September 27 at 23:59 UTC.
💻 Kimi Code: launched a desktop app for macOS and Windows that manages multiple agents in one space and runs tasks in parallel on long-horizon work.
🎞️ Motion: added Motion Studio, a visual editor for animations on a live site, plus the Ultramotion editing agent, with inspection and visual editing free.
🛡️ Aikido: released Altar-1, its first open-weight security model, built for defensive work and meant to be deployed on your own infrastructure.
🤖 Devin: added Devin Cloud in Terminal, so /cloud creates and resumes sessions from the CLI, and SSH access to Devin's VM with /handoff to your device.
THAT’S IT FOR TODAY
Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!
See you tomorrow :)
- Dr. Alvaro Cintas
✓ Full archive of premium guides with ready-to-use prompts
✓ Structured AI courses (step-by-step, start-to-finish)
✓ Every upcoming premium tutorial